Re: xfs oops (CVS-2004-05-15_05:00_UTC)

[Christoph Hellwig <hch@xxxxxxxxxxxxx>]

> Okay, I've run even more tests yesterday and during the night, conclusions:
> 
> - current CVS without 1.1371.750.18 - oops in ... minutes
> - current CVS without undelay changes
>   (xfs_buf.c 1.171, xfs_buf.h 1.97) - minutes.. aswell
> - current CVS without 1.1371.750.18 _and_ undelay changes - no problems
>   (tested 10 hours)
> - current CVS without 1.1371.750.18, 1.1587.5.6 1.1722.10.100 and undelay
>   changes - no problems (tested 14 hours)

Hmmm, this makes me wonder whether we might still have a freed buffer
on the delwri list somewhere.  Can you test with the patch below instead
of the previous one?


===== fs/xfs/linux-2.6/xfs_buf.c 1.116 vs edited =====
--- 1.116/fs/xfs/linux-2.6/xfs_buf.c    2004-05-28 03:45:07 +02:00
+++ edited/fs/xfs/linux-2.6/xfs_buf.c   2004-06-24 14:13:12 +02:00
@@ -310,7 +310,15 @@
 {
        PB_TRACE(bp, "free", 0);
 
+       ASSERT(list_empty(&pb->pb_list));
        ASSERT(list_empty(&bp->pb_hash_list));
+       ASSERT(!(bp->pb_flags & PBF_DELWRI));
+
+       if (!bp->pb_pages) {
+               printk("pagebuf without pb_pages?\n");
+               dump_stack();
+               goto out;
+       }
 
        if (bp->pb_flags & _PBF_PAGE_CACHE) {
                uint            i;
@@ -331,6 +339,7 @@
                _pagebuf_free_pages(bp);
        }
 
+ out:
        pagebuf_deallocate(bp);
 }